468 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND …roman/professional/... · Index Terms—Biosensor, DNA, electrochemical sensor, fre-quency response analysis, impedance spectroscopy,

468 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 6, NO. 5, OCTOBER 2012

16-Channel CMOS Impedance Spectroscopy DNAAnalyzer With Dual-Slope Multiplying ADCs

Hamed Mazhab-Jafari, Student Member, IEEE, Leyla Soleymani, and Roman Genov, Senior Member, IEEE

Abstract—We present a 16-channel, mixed-signal CMOS DNAanalyzer that utilizes frequency response analysis (FRA) to extractthe real and imaginary impedance components of the biosensor.Two computationally intensive operations, the multiplication andintegration required by the FRA algorithm, are performed byan in-channel dual-slope multiplying ADC in the mixed-signaldomain resulting in minimal area and power consumption.Multiplication of the input current by a digital coefficient isimplemented by modulating the counter-controlled duration ofthe charging phase of the ADC. Integration is implemented byaccumulating output digital bits in the ADC counter over multipleinput samples. The 1.05 mm 1.6 mm prototype fabricated ina 0.13 m standard CMOS technology has been validated inprostate cancer DNA detection. Each channel occupies an area ofonly 0.06 mm and consumes 42 W of power from a 1.2 V supply.

Index Terms—Biosensor, DNA, electrochemical sensor, fre-quency response analysis, impedance spectroscopy, multiplyingADC.

I. INTRODUCTION

A NUMBER of integrated circuits for DNA detection ona disposable CMOS substrate platform have been intro-

duced in recent years [1]–[7]. The on-chip electrochemical am-perometric sensing DNA detection method shows promise overoptical methods due to the real-time nature of DNA analysis[8]. Fig. 1 depicts a basic electrochemical cell employed in am-perometric sensing. It consists of a working electrode (WE)and a reference electrode (RE). A transducer holds the workingelectrode at a known potential while recording the redoxcurrent generated as a result of the voltage difference betweenthe two electrodes. The reference electrode potential isset to either a constant voltage for constant-potential amperom-etry, a bidirectional ramp voltage for cyclic voltammetry, or asmall-amplitude sinusoid for impedance spectroscopy.In biochemical sensors, biologically active compounds such

as artificial antibodies, enzymes or receptors are integrated ontothe working electrode. The integration process is performedby immobilization and stabilization of biological molecules

Manuscript received March 16, 2012; revised June 28, 2012 and September03, 2012; accepted September 28, 2012. Date of publication November 20,2012; date of current version November 28, 2012. This paper was recommendedby Associate Editor P. Hafliger.H. Mazhab-Jafari and R. Genov are with the Department of Electrical and

Computer Engineering, University of Toronto, Toronto, ON M5S 3G4, Canada(e-mail: [email protected]; [email protected]).L. Soleymani is with McMaster University, Hamilton, ON L8S 4L8, Canada

(e-mail: [email protected]).Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TBCAS.2012.2226334

Fig. 1. Conceptual view of a biosensor impedance spectroscopy system.

on the electrode surface. In DNA sensing applications, thesurface of the working electrode is functionalized with a probeDNA. Binding of the probe DNA with the target DNA resultsin a variation of the working electrode surface properties suchas the impedance or the surface charge. The variation of thesurface properties results in a change in the waveform featuresof the redox current, thus indicating changes in the kineticsand thermodynamics of the chemical reactions at the sensoryinterface [9]–[11]. In most applications, the recorded redoxcurrent is in the range of pA to nA and the biosensor frequencyresponse covers a range of 1 Hz to 10 kHz [13]–[17].Impedance spectroscopy is a popular method of quantitative

and qualitative monitoring of chemical reactions in manybiosensors. A wide range of biosensors have been developedwhich rely on impedance spectroscopy, including sensors fordetection of enzymes, antibodies and DNA [19], [20]. Fig. 1also shows the basic principle of electrochemical impedancespectroscopy in DNA sensing applications. Binding of theprobe DNA with the target DNA changes the impedance be-tween the reference and the working electrodes. To detect thischange, an impedance spectrum analyzer extracts the real andimaginary components of the biosensor impedance. A generic

biosensor impedance model is depicted in Fig. 1. Inthis model represents the electrolyte resistance between theworking and reference electrode, represents the diffusionlayer capacitance, models the interfacial double-layercapacitance at the WE-electrolyte interface and modelsthe charge transfer resistance at the WE-electrolyte interface[21].Several impedance spectroscopy integrated circuits have

been reported [21]–[23]. The design in [22] is a single-channelimpedance extractor based on a lock-in amplifier that extractsthe sensor impedance from 1 Hz to 10 kHz. The implemen-tation in [21] is a 100-channel impedance-to-digital converterbased on a delta-sigma modulator capable of extracting sensorimpedance from 1 mHz to 10 kHz at the expense of a long

1932-4545/$31.00 © 2012 IEEE

MAZHAB-JAFARI et al.: 16-CHANNEL CMOS IMPEDANCE SPECTROSCOPY DNA ANALYZER 469

conversion cycle. A direct conversion receiver without anon-chip ADC [23] extracts the electrode impedance from 10Hz to 50 MHz at the cost of consuming 104 mW of power.This paper presents a scalable, multi-channel, compact and

low-power CMOS impedance spectroscopy DNA analyzer.Frequency response analysis (FRA) algorithm is utilized toextract the real and imaginary components of the biosensorimpedance [24]. The proposed microsystem consists of aprogramable on-chip waveform generator and 16 impedanceextraction channels. Each channel includes a current-modeinput dual-slope multiplying ADC. It efficiently performsmultiplication and integration, two computationally inten-sive operations required to implement the FRA algorithm.Multiplication of the input current by a digital coefficient isimplemented by modulating the counter-controlled duration ofthe charging phase of the ADC by that coefficient. Integrationis implemented by accumulating the output digital bits in theADC counter. The dual-slope multiplying ADC utilizes mostlythe same circuits as a conventional dual-slope ADC, and themultiplication and integration are achieved by modifying theADC algorithm. The rest of the paper is organized as follows.Section II presents the principle of DNA detection on a CMOSdie. Section III presents the impedance spectroscopy VLSIarchitecture. Section IV details the circuit implementation ofthe VLSI architecture. Section V demonstrates the electricalexperimental results obtained from a 0.13 m CMOS proto-type. In Section VI, results of on-chip impedance spectroscopyof DNA in prostate cancer screening are presented.

II. DNA DETECTION PRINCIPLE

The principle of the label-free DNA detection methodis shown in Fig. 2. It employs potassium ferrocyanideK [Fe(CN )] reporter. Potassium ferrocyanide is a nega-tively charged redox complex with well-defined electrochem-ical signature exhibiting oxidation and reduction currents at

of mV and mV respectively [25].The maximum electron transfer between the electrode andpotassium ferrocyanide is achieved in the absence of DNAtarget and probe as illustrated in Fig. 2(a). The oxidizationcurrent drops when the Au electrode surface is hybridizedwith negatively charged probe DNA, as illustrated in Fig. 2(b).When the complementary DNA binds with the probe DNA, thesurface negative charge further increases and the oxidizationcurrent is further reduced, as shown in Fig. 2(c). This change inthe current is an indicator of the target DNA presence and con-centration. In other words, the presence of negatively chargedDNA on the biosensor surface is translated to a decrease in thepotassium ferrocyanide oxidation/reduction current creating adetectible signal change [26], [27].The DNA sensing electrodes are created by first forming

passivation openings on the top metal layer (aluminum) ofthe CMOS die similarly to how it is done for wire bond pads.Electroless electroplating is then employed to deposit nickel(2 m), palladium (0.2 m) and gold (0.1 m) on the exposedAl surface to form a bio-compatible electrode surface. After theelectrode fabrication, the die is wire-bonded and the bondingwires are insulated with a biocompatible epoxy to enable

Fig. 2. Label-free electrochemical DNA detection principle.

Fig. 3. Block diagram of a frequency-response analyzer (FRA) system forbiosensor impedance spectroscopy.

on-chip electrochemical experiments without damaging thebonding wires [18].

III. IMPEDANCE SPECTROSCOPY VLSI ARCHITECTURE

A small-signal model of the electrode-electrolyte interfacein an electrochemical cell is shown in the center of Fig. 1.Fast fourier transform (FFT) and frequency-response analyzer(FRA) are two methods widely used for characterizing theelectrode impedance [24]. Compared to the FFT, the FRAmethod requires simple circuitry and can be implemented ona small silicon area, making it suitable for sensory array mi-crosystems. A functional block diagram of the FRA algorithmis shown in Fig. 3. The sensor is interrogated with a sinusoidalvoltage. Multiplication of the sensor responsewith or and integration of the results over onecycle of the interrogation signal results in separation of thereal and the imaginary components of the sensor impedance asfollows [24]

(1)

(2)

where represents the amplitude, represents the phase ofthe sensor impedance and is the period of the interrogation


Fig. 4. Impedance spectroscopy microsystem functional block diagram.

signal. is the real portion and is the imag-inary portion of the sensor impedance. These values over awide frequency range are sufficient to fully describe the sensorimpedance.In this work the FRA algorithm has been chosen to implement

a sensory array impedance spectroscopy microsystem. The twokey components in this system are the multiplier and the digitalintegrator. Both of these operations are implemented with anin-channel multiplying dual-slope ADC that reuses the circuitsof a conventional dual-slope ADC [28].

IV. CIRCUIT IMPLEMENTATION

A. Multi-Channel System-Level Architecture

The functional block diagram of the impedance spectroscopymicrosystem based on the FRA algorithm is shown in Fig. 4.The microsystem is comprised of a programable analog wave-form generator, a programmable digital pattern generatorand an array of impedance extraction units. The waveformgenerator produces the interrogation waveform anddrives the reference electrode with it. The digital pattern gen-erator generates digital multiplication coefficients representingeither sine wave or cosine wave that are synchronized withthe interrogation waveform. Each impedance extraction unitconsists of a dual-slope multiplying ADC (DS-MADC). Thefront-end current-to-voltage converter is an analog integratorthat acquires an input current at the low-impedance input nodeset to a controlled potential. The DS-MADC multiplies thebiosensor response with a set of digital coefficientsrepresenting or that are synchronized withthe analog sinusoidal interrogation voltage on the referenceelectrode. Next the DS-MADC accumulates the results overone period of the interrogation signal using a digital integrator(counter), thus extracting the real or the imaginary componentsof the biosensor impedance.

B. Waveform Generator and Pattern Generator

The frequency response analysis implementation is the sim-plest when rectangular waveforms are used instead of sine wavefor both the interrogation and the multiplication signals. Theproblem is that severe systematic errors appear due to the higherorder harmonics existing in the rectangular waveforms. Step-wise approximation of the interrogation waveform and the mul-tiplication signals reduces the effect of higher order harmonics

Fig. 5. Waveform generator functional block diagram.

and increases the measurement accuracy. It has been shown [22]that representing both the interrogation signal and the multipli-cation signals by a coarsely quantized approximation can sig-nificantly reduces the error due to the higher order harmonicsand can significantly reduce the measurement inaccuracy.In this work a programmable analog waveform generator is

utilized to generate the stepwise approximation of the interro-gation signal. The block diagram of the programmable wave-form generator is shown in Fig. 5. It is composed of an 8-bitR-2R DAC, an on-chip SRAM1 and a bidirectional counter.The DAC coefficients are stored in the on-chip SRAM1. Theon-chip waveform generator generates both stepwise sinusoidaland cyclic voltammetry (CV) ramp waveforms. It provides awide range of user-controlled rate and amplitude parameterswith a maximum CV scan range of 1.2 V, and scan rate rangingbetween 0.1 mV/sec to 400 V/sec and the sinusoidal frequencyrange from 100 mHz to 50 kHz. An off-chip clock generatorprovides the variable frequency clock signal enabling the wave-form generator to generate the interrogation signal at differentfrequencies. The DAC occupies an area of 0.012 mm and dis-sipates 1.1 mW of power from a 1.2 V supply when driving aload of 5 nF at 50 kHz. At low frequencies the interrogationsine wave is represented by 64 samples per period. The DACcoefficients for the first 16 samples ( to ) of the wave-form generator output waveform and for the corresponding dig-ital sin/cos multiplication coefficients are stored in two on-chipglobal SRAM banks (SRAM1 and SRAM2, respectively), asshown in Fig. 4. The SRAMs occupy an area of 0.028 mm anddissipate 0.9 W of power when clocked at 50 kHz. By sym-metry, 64 samples in one period are generated from the 16 sam-ples stored on-chip. As shown in Fig. 6, for a sine wave, the first16 samples stored in the SRAM1 generates the first quadrant ofthe sine wave. In this case the counter controlling the SRAM1is counting up and is set to 00, in Fig. 5. In the next quadrantthe counter controlling the SRAM1 counts down thus reversingthe order of the samples and generating the second quadrant ofthe sine wave. is set to 00 in this case. The third and forthquadrants are generated in the same manner but in this case thepolarities of the samples fed into the DAC are reversed (by set-ting to 01). To generate a cosine, the cycle starts from thesecond quadrant instate of the first one. If a CV waveform isrequired the up/down counter is directly interfaced to the wave-form generator DAC (by setting to 11).As the interrogation frequency increases, the number of sam-

ples representing the interrogation and multiplication signalsdecreases and at 10 kHz both signals are represented by three


Fig. 6. Principle of the sine wave generation.

samples. This greatly reduces the ADC speed requirement whilethe error caused by the reduction in number of samples is keptlow by averaging the results over multiple cycles.An ideal model of the impedance spectroscopy microsystem

has been constructed using verilog-A model components (idealanalog integrator and multiplying ADC) to study the effectof step-wise approximation of the interrogation signal on theoverall system accuracy. First a simulation was performedwhere a sinusoidal voltage stimulus (8-bit accurate generatedby an ideal DAC) with the frequency swept from 0.1 Hz to10 kHz was applied to the biosensor model shown in Fig. 4.The value of was set to 1 G was set to 500 pF,

was set to 300 pF and was set to 1 M . The sensorresponse was recorded with the ideal analog integrator andmultiplied, using an ideal multiplying dual-slope ADC, by an8-bit accurate digital multiplication coefficient representingsine or cosine in the FRA algorithm in Fig. 3. Next, thesame set of simulations were performed in which both digitalmultiplication coefficients and the analog sinusoidal voltageinterrogation signal are represented by 64 samples per periodat low frequencies. In this simulation as the interrogationfrequency increases, the number of samples representing theinterrogation and multiplication signals decreases. Also, inthis simulation the results are averaged over multiple cycles

to reduce the errors caused by reduction in the numberof samples. The absolute value of the relative error of thebiosensor impedance computed using 8-bit accurate interroga-tion signal and 8-bit accurate multiplication coefficients versusthe simulated biosensor impedance obtained from stepwiseapproximation of these signals is shown in Fig. 7. The absolutevalue of the relative error stays below 0.9% and 1.0% for realand imaginary component of the biosensor, respectively.

C. Dual-Slope Multiplying ADC Channel

The VLSI architecture of one channel of the integrated spec-trum analyzer is depicted in Fig. 8(a). Each channel consists ofan integrating amplifier with an on-chip 10 pF capacitor ,a high-speed latched comparator and digital blocks. The inte-grator switches are implemented with low-leakage switches asshown.The conventional dual-slope ADC operates in two phases as

depicted in Fig. 9. In phase I the integrating capacitor ischarged for a predetermined period of time . Next, during the

Fig. 7. Absolute value of the relative error of the biosensor modelimpedance as a function of frequency due to stepwise approximation of the in-terrogation signal for the (a) real and (b) imaginary components.

second phase of the operation, the capacitor is discharged to zeroby a DC reference current. By counting the time , a digitalrepresentation of can thus be obtained as .To implement multiplication of the input current by a digitalsin/cos coefficient ( to ) as needed by the FRA algo-rithm, the duration of phase I is scaled with a constant coeffi-cient as shown in Fig. 9. In this case by counting thetime , a digital representation of can be obtained as

. To extract the real and imaginary compo-nents of the biosensor impedance, the input current is multi-plied by the reference sine or cosine digital coefficient denotedas (stored in SRAM2, in Fig. 4) and the results are integratedover one period by a 16-bit counter as follows:

(3)

(4)


Fig. 8. (a) Dual-slope multiplying ADC VLSI architecture. (b) Timing diagram of all relevant signals.

Fig. 9. Timing diagram illustrating the ADC multiplication function.

Fig. 10. Chopper-stabilized folded-cascode OTA1 in the analog integrator.

where is the number of samples in one period andand are the digital multiplication coefficients that arestored in the SRAM2. The multiplying dual slope ADC per-forms the multiplication required by (3) and (4), thus the needfor 16 digital multipliers is eliminated.The timing diagram of the ADC for a typical conversion

cycle is shown in Fig. 8(b). First, the integrating counter isreset. At the same time, the sin/cos multiplication coefficient,, is loaded into the in-channel input latch (time A). Next, the

in-channel counter counts up from zero to time and theinput current is integrated onto capacitor . After time ,the voltage on the capacitor is held constant for a fixed timeinterval . During this time the content of the output latch(zero for the first conversion cycle) is loaded into the counter(time B). During time , depending on the comparator

Fig. 11. Comparator circuit diagram. (a) One of the three identical gain stages.(b) The high-speed latch.

Fig. 12. Simulated input-referred noise spectrum of the analog integrator withthe chopper on and off (clocked at 100 kHz) from 0.01 Hz to 10 kHz.

output, the integrating capacitor is discharged using the appro-priate current source, or . During time thecounter counts up or down depending on the sign of the inputcurrent in phase I and the final value of the counter is writteninto the output latch (time C). This part of the ADC conversioncycle performs the summation which implements integrationrequired by the FRA algorithm. This process is repeated forone full cycle of the sinusoidal stimulation waveform and thefinal value stored in the output latch corresponds to the real orimaginary component of the biosensor impedance, for sin andcos multipliers respectively.


Fig. 13. (a) Circuit architecture of the three-electrode regulation loop. (b) OTA2 and OTA3 circuit schematic.

TABLE IOTA1 AND COMPARATOR TRANSISTOR SIZING

The analog integrator amplifier shown in Fig. 10 is a folded-cascode transconductance amplifier consuming 6 W from a1.2 V supply. Impedance spectroscopy is performed over thefrequency range of 1 Hz to 10 kHz. To reduce the effect offlicker noise, the amplifier utilizes input PMOS devices with ahigh aspect ratio and internal chopping. As shown in Fig. 10, thechopper switches are placed at the input of the OTA. Anotherset is placed between the cascoded NMOS current mirrors. Thissignificantly reduces the flicker noise and DC offsets caused bythe input pair transistors and the NMOS current mirror transis-tors.The simulated input-referred noise of the integrator for the

cases where the chopper is disabled and enabled is shown inFig. 12. The integrated input-referred noise from 0.01 Hz to10 kHz is 0.12 pA when the chopper is disabled and 0.07 pAwhen the chopper is enabled. The integrator transistor sizes areshown in Table I. Contribution of each transistor to the totalinput-referred noise is shown Table II. When the chopper is dis-abled the main contributions are from the OTA current mirrortransistors and and the input pair transistors and .When the chopper is enabled, the current mirror transistorsand are the main contributors to the input-referred noise.The comparator is implemented with three pre-amplifier

stages [Fig. 11(a)], with a total gain of 60 dB, and a high-speedoutput latch as shown in Fig. 11(b). The comparator is verifiedin simulation to operate at up to 40 MHz. The full channelconsumes 42 W of power from a 1.2 V supply when clocked at10 MHz. The comparator transistor sizes are shown in Table I.

TABLE IIANALOG INTEGRATOR NOISE SUMMARY

TABLE IIIOTA2 AND OTA3 TRANSISTORS SIZING

D. Three-Electrode Regulation Loop

The functional block diagram of the regulation loop is shownin Fig. 13(a). A three-electrode potentiostat configuration [29]is employed as opposed to the basic two-electrode configura-tion shown in Fig. 1. In a two-electrode configuration the cur-rent flowing between the two electrodes alters the voltage dropacross due to the non-zero impedance of the electrolyte so-lution and the electrode-electrolyte interface. In the three-elec-trode configuration, a third, counter electrode (CE), rather thanthe reference electrode is placed in the current path. The refer-ence electrode does not sink/source an electric current. Thus thevoltage difference between the working and reference electrodeis maintained at a known voltage independent of the electro-chemical cell impedance. The three-electrode regulation loop iscomposed of the two OTAs as shown in Fig. 13(a). To allowfor wide input and output swings class-AB OTAs with bothNMOS and PMOS input differential pairs are used for the po-tentiostat as shown in Fig. 13(b). OTA transistor sizing is given


Fig. 14. Die micrograph of the 1.05 mm 1.6 mm 16-channel integratedimpedance spectrum analyzer fabricated in a 0.13 m standard CMOS tech-nology.

Fig. 15. Experimentally measured transfer characteristics of the impedancespectrum analyzer channel for three sampling frequencies.

Fig. 16. Experimentally measured output spectrum of the ADC for a 10 Hzsinusoidal input.

in Table III. Assuming that is fixed by the analog inte-grator, the regulation loop configuration ensures that

tracks the voltage difference between the output of thewaveform generator and the positive terminal of the OTA3.

Fig. 17. Waveform generator DNL.

Fig. 18. Waveform generator INL.

Fig. 19. Total harmonic distortion of waveform generator DAC.

V. ELECTRICAL EXPERIMENTAL RESULTS

The 16-channel 1.2 1.6 mm integrated impedance spec-trum analyzer prototype was implemented in a 0.13 m CMOSprocess with a 1.2 V supply. The die micrograph is shown inFig. 14. Each channel consists of 4WEs, with differentWE sizes(55 m 55 m, 5 m 5 m, and two 2 m 2 m) to covera wide combined dynamic range. Each column of four workingelectrodes shares a 45 m-wide Au reference and counter elec-trodes driven by the on-chip waveform generator. Each channelconsists of the analog integrator, dual-slope multiplying ADCand in-channel bias and clock generator. Each channel WE can


TABLE IVEXPERIMENTALLY MEASURED CHARACTERISTICS

Fig. 20. Off-chip biosensor model impedance as a function of frequency ex-perimentally measured by the impedance spectroscopy microsystem. (a) Real.(b) Imaginary components.

be set to a specific voltage independently, allowing for simulta-neous detection of different DNA strands.The digital output of one channel for the input current swept

between 100 fA and 400 nA is shown in Fig. 15. The static inputdynamic range is 140 dB cumulatively for the three samplingfrequency settings, or 55.4 dB at 1 MHz clock. The dynamicrange for each setting is computed by taking the ratio of themaximum signal that saturates the ADC to the LSB for a givensampling frequency setting. Dynamic performance of an entirechannel was measured by applying a 10 Hz full scale (400 nA)

Fig. 21. Absolute value of the relative error of the off-chip biosensor modelimpedance as a function of frequency. (a) Real. (b) Imaginary components.

sinusoidal input current with the ADC clocked at 10 MHz.Fig. 16 shows the 65536-point FFT of the measured ADCoutput. The strong second harmonic is due to the single-endedarchitecture of the ADC. The resulting effective number of bits(ENOB) is 9.3. The DNL and INL of the waveform generatorare shown in Figs. 17 and 18, respectively. The waveformgenerator achieves an static accuracy of 8 bits. Dynamic per-formance of the waveform generator was characterized bymeasuring the total harmonic distortion of the DAC versusfrequency. Fig. 19 shows the THD of the measured DACoutput. The DAC archives a THD of greater than 42 dB overthe operating frequency range.The electrode model shown in Fig. 1 was used first to emulate

the biosensor. The value of was set to 1 G was setto 500 pF, was set to 300 pF and was set to 1 M . Toverify the impedance extraction capability of the microsystem,a sinusoidal voltage stimulus (generated by the on-chip DAC)with the frequency swept from 0.1 Hz to 10 kHz was appliedto the biosensor model. Fig. 20(a) and (b) demonstrate that thefabricated prototype tracks the theoretical model well over thefull range of frequencies. The absolute value of the relative errorof the ideal biosensor impedance verses the measured biosensorimpedance as a function of frequency is shown in Fig. 21. Theabsolute value of the relative error stays below 8.4% and 7.5%for real and imaginary component of the biosensor respectively.Table IV provides a summary of experimentally measured

characteristics of the integrated impedance spectroscopy mi-crosystem.


Fig. 22. Potassium ferricyanide solution impedance as a function of frequencyexperimentallymeasured by the impedance spectroscopymicrosystem. (a) Real.(b) Imaginary components.

VI. ELECTROCHEMICAL EXPERIMENTAL RESULTS

A set of electrochemical experiments have been conductedto validate the functionality of the impedance spectroscopy mi-crosystem. In these tests a 3-electrode setup has been utilizedwith on-chip gold-plated 55 m 55 m working electrode,on-chip gold plated counter electrode and an off-chip Ag-AgClreference electrode. The reference and counter electrodes aredriven by the on-chip three-electrode regulation loop as shownin Fig. 13(a) and all the excitation waveforms are generatedusing the on-chip waveform generator.First, in order to validate the design, impedance spectroscopy

recordings of 0.1 mM and 1mMpotassium ferricyanide in 1 Mpotassium phosphate buffer (pH 7.3) were carried out. A 9 mV0.1 Hz to 10 kHz sine wavewas applied between theWEs and anoff-chip Ag-AgCl reference electrode. The real and imaginaryimpedance results obtained from the two concentrations of thepotassium ferricyanide solution are shown in Figs. 22(a) and (b).An increase in the concentration of the potassium ferricyanideresults in a decrease in value of and and increase invalue of . The measurements of potassium ferricyanidevalidate impedance spectroscopy microsystem in DNA sensingapplications.The SoC was also extensively validated in DNA analysis for

detection of prostate cancer. Fig. 23(a) shows real and imag-inary impedance components of an on-chip Au electrode hy-bridized with 5 M single-stranded prostate cancer probe DNA.

Fig. 23. (a) Impedance spectrum of 5 M prostate cancer DNA probe, comple-mentary, and non-complementary targets. (b) The corresponding 3-sigma errorbars (from 3 chips, 20 measurements each).

Adding a 5 M noncomplementary target DNA does not signif-icantly change the real and imaginary components of the elec-trode impedance indicating that non-specific adsorption is neg-ligible. Adding a 5 M complementary prostate cancer targetDNA leads to significant reduction of potassium ferrocyanideredox current, resulting in increase in both real and imaginarycomponents of the electrode impedance. This is due to the ad-ditional negative surface charge resulting from formation ofdouble-stranded DNA on the electrode surface. The 3-sigmaerror bars (from 3 chips) with the real and imaginary impedancedetection noise margins of approximately 18.3 K and 20.9Mrespectively are shown in Fig. 23(b).Table V provides a comparative analysis of the presented de-

sign and existing amperometric biochemical sensory microsys-tems. The design presented in this work achieves the highestdynamic range and the lowest sensitivity in terms of LSB. Also,the design presented in this paper requires less time to extractreal and imaginary components of the biosensor compared to


TABLE VCOMPARATIVE ANALYSIS OF REPORTED AMPROMETRIC SENSORY MICROSYSTEM

the FRA implementations in which a square waveform is uti-lized for multiplication and integration. The stepwise approx-imation of the interrogation waveform and multiplication co-efficients significantly reduces the systematic error caused byhigher order harmonics present in the rectangular waveforms,thus eliminating the need for excessive averaging that is timeconsuming at the lower frequency range. This enables real-timemonitoring and analysis of DNA hybridization.

VII. CONCLUSION

A 16-channel, mixed-signal CMOS impedance spectroscopyDNA analyzer is presented. It consists of a programable wave-form generator, on-chip memory and multiple impedance ex-traction units. Multiplication and integration, two operations re-quired for frequency response analysis (FRA) algorithm, are ef-ficiently performed by the in-channel current-mode input dual-slope multiplying ADCwith negligible resources overhead. TheADC combines impedance extraction and analog-to-digital con-version into a single conversion cycle. The impedance spec-troscopy microsystem was implemented in a CMOS 0.13 mtechnology. Each channel occupies an area of 0.06 mm andconsumes 42 W of power from a 1.2 V supply.

REFERENCES

[1] F. Heer, M. Keller, G. Yu, J. Janata, M. Josowicz, and A. Hierlemann,“CMOS electro-chemical DNA-detection array with on-chip ADC,” inProc. IEEE Int. Solid-State Circuits Conf., 2008, pp. 168–169.

[2] M. Schienle, C. Paulus, A. Frey, F. Hofmann, B. Holzapfl, P. S. Bauer,and R. Thewes, “A fully electronic DNA sensor with 128 positions andin-pixel A/D conversion,” IEEE J. Solid-State Circuits, vol. 39, no. 12,pp. 2438–2445, 2004.

[3] D. M. Garner, H. Bai, P. Georgiou, T. G. Constandinou, S. Reed, L. M.Shepherd, W. Wong, K. T. Lim, and C. Toumazou, “A multichannelDNA SoC for rapid point-of-care gene detection,” in Proc. IEEE Int.Solid-State Circuits Conf., 2010, pp. 492–493.

[4] A. Hassibi and T. H. Lee, “A programmable 0.18 m CMOS electro-chemical sensor microarray for bimolecular detection,” IEEE SensorsJ., vol. 6, pp. 1380–1388, 2006.

[5] A. Hassibi, H. Vikalo, J. L. Riechmann, and B. Hassibi, “Real-timeDNA microarray analysis,” Nucl. Acids Res., pp. 1–12, Aug. 2009.

[6] B. Jang, P. Cao, A. Chevalier, A. Ellington, and A. Hassibi, “A CMOSfluorescence-based biosensor microarray,” in Proc. IEEE Int. Solid-State Circuits Conf., Feb. 2009, pp. 436–437.

[7] H. Wang, Y. Chen, A. Hassibi, A. Scherer, and A. Hajimiri, “A fre-quency-shift CMOS magnetic biosensor array with single-bead sensi-tivity and no external magnet,” in Proc. IEEE Int. Solid-State CircuitsConf., Feb. 2009, pp. 438–439.

[8] P. M. Levine, P. Gong, R. Levicky, and K. L. Shepard, “Active CMOSsensor array for electrochemical biomolecular detection,” IEEE J.Solid-State Circuits, vol. 43, no. 8, pp. 1859–1871, 2008.

[9] J.Wang, “Electrochemical nucleic acid biosensors,”Analytica ChimicaActa, vol. 469, pp. 63–71, 2002.

[10] T. G. Drummond, M. G. Hill, and J. K. Barton, “ElectrochemicalDNA sensors,” Nature Biotechnol., vol. 21, no. 10, pp. 1192–1199,Oct. 2003.

[11] D. A. D. Giusto, W. A. Wlassoff, S. Giesebrecht, J. J. Gooding, and G.C. King, “Enzymatic synthesis of redox-labeled RNA and dual-poten-tial detection at DNA-modified electrodes,” Angewandte Chemie Int.Ed., vol. 43, pp. 2809–2812, 2004.

[12] M. Schena, D. Shalon, R. W. Davis, and P. O. Brown, “Quantitativemonitoring of gene expression patterns with a complementary DNAmicroarray,” Science, vol. 270, no. 5235, pp. 467–470, Oct. 1995.

[13] C. Yang, Y. Huang, B. L. Hassler, R. M. Worden, and A. J. Mason,“Amperometric electrochemical microsystem for a miniaturized pro-tein biosensor array,” IEEE Trans. Biomed. Circuits Syst., vol. 3, no. 3,pp. 160–168, 2009.

[14] R. Genov, M. Stanacevic, M. Naware, G. Cauwenberghs, and N.Thakor, “16-channel integrated potentiostat for distributed neuro-chemical sensing,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 53,no. 11, pp. 2371–2376, Nov. 2006.

[15] M. Roham,M. P. Garris, and P.Mohseni, “Awireless IC for time-sharechemical and electrical neural recording,” IEEE J. Solid-State Circuits,vol. 44, no. 12, pp. 3645–3658, 2009.

[16] M. Stanacevic, K. Murari, A. Rege, G. Cauwenberghs, and N. Thakor,“VLSI potentiostat array with oversampling gain modulation for wide-range neurotransmitter sensing,” IEEE Trans. Biomed. Circuits Syst.,vol. 1, no. 1, pp. 63–72, Feb. 2007.

[17] M. Nazari and R. Genov, “A fully differential CMOS potentiostat,”in Proc. IEEE Int. Symp. Circuits and Systems, May 2009, vol. 2, pp.2177–2180.


[18] M. Nazari, H. M. Jafari, and R. Genov, “192-channel CMOS neuro-chemical microarray,” in proc. IEEE Custom Integrated Circuits Conf.,Sep. 2010, pp. 121–125.

[19] J. Kafka, O. Pnke, B. Abendroth, and F. Lisdat, “A label-free DNAsensor based on impedance spectroscopy,” Electrochimica Acta, vol.53, no. 25, pp. 7467–7474, Oct. 2008.

[20] Y. T. Long, C. Z. Li, H. B. Kraatz, and J. S. Lee, “AC impedancespectroscopy of native DNA and M-DNA,” Biophys. J., vol. 84, no.5, pp. 3218–3225, May 2003.

[21] C. Yang, S. R. Jadhav, R. M. Worden, and A. J. Mason, “Compact low-power impedance-to-digital converter for sensor array microsystems,”IEEE J. Solid-State Circuits, vol. 44, no. 10, pp. 2844–2855, 2009.

[22] M.Min and T. Parve, “Improvement of lock-in bio-impedance analyzerfor implantable medical devices,” IEEE Trans. Instrum. Meas., vol. 56,pp. 968–974, 2007.

[23] A. Manickam, A. Chevalier, M. McDermortt, A. D. Ellington, and A.Hassibi, “A CMOS electrochemical impedance spectroscopy biosensorarray for label-free biomolecular detection,” in Proc. IEEE Int. Solid-State Circuits Conf., 2010, pp. 130–131.

[24] E. Barsoukov and J. R. Macdonald, Impedance Spectroscopy: Theory,Experiment, and Applications, 2nd ed. Hoboken, NJ: Wiley, 2005.

[25] M. Alubaidy, L. Soleymani, K. Venkatakrishnan, and B. Tan, “Fem-tosecond laser nanostructuring for femtosensitive DNA detection,”Biosens. Bioelectron., vol. 33, no. 1, pp. 83–87, Mar. 2012.

[26] L. Aorga, A. Rebollo, J. Herrn, S. Arana, E. Bandrs, and J. Garca-Fon-cillas, “Development of a DNA microelectrochemical biosensor forCEACAM5 detection,” IEEE Sensors J., vol. 10, no. 8, pp. 1368–1374,Aug. 2010.

[27] Y. S. Choi, K. S. Lee, and D. H. Park, “A genomic detection using anindicator-free DNA on a DNA chip microarray,” Appl. Phys., vol. 6,pp. 772–776, 2006.

[28] H. M. Jafari and R. Genov, “CMOS impedance spectrum analyzer withdual-slope multiplying ADC,” in Proc. IEEE Biomedical Circuits andSystems Conf., San Diego, CA, Nov. 2011.

[29] L. Li, X. Liu, W. A. Qureshi, and A. J. Mason, “CMOS amperometricinstrumentation and packaging for biosensor array applications,” IEEETrans. Biomed. Circuits Syst., vol. 5, no. 5, pp. 439–448, 2011.

Hamed Mazhab-Jafari (S’11) received the B.Eng.and M.A.Sc. degrees in electrical engineering fromMcMaster University, Hamilton, ON, Canada, in2004 and 2006, respectively.He is currently working toward the Ph.D. degree in

electrical and computer engineering at the Universityof Toronto, Toronto, ON, Canada. His M.A.Sc. thesisfocused on low-power ultra-wideband CMOS frontend and ultra-widebands antennas. His Ph.D disser-tation focus is on CMOS DNA analysis SoC. He hasheld internship positions at Kapik Integration, where

he worked on low-power mixed signal circuits. Since 2011, he has been withSnowbush IP, Toronto, ON, Canada, where he focuses on research and devel-opment of next-generation high-speed wireline communication systems.

Leyla Soleymani received the Ph.D. degree in elec-trical and computer engineering from the Universityof Toronto, Toronto, ON, Canada, in 2010.She is currently an Assistant Professor in the

Department of Engineering Physics and Schoolof Biomedical Engineering, McMaster University,Hamilton, ON, Canada. Her research interestsinclude development of new hierarchical materialsand combining different micro/nanofabricationtechnologies for building chip-based biosensors.

Roman Genov (S’96–M’02–SM’11) received theB.S. degree (first rank) in electrical engineering fromthe Rochester Institute of Technology, Rochester,NY, in 1996, and the M.S. and Ph.D. degrees inelectrical and computer engineering from JohnsHopkins University, Baltimore, MD, in 1998 and2002, respectively.He held engineering positions at Atmel Corpora-

tion, Columbia, MD, in 1995 and Xerox Corporation,Rochester, NY, in 1996. He was a Visiting Researcherin the Laboratory of Intelligent Systems at the Swiss

Federal Institute of Technology (EPFL), Lausanne, Switzerland, in 1998 andin the Center for Biological and Computational Learning at the MassachusettsInstitute of Technology, Cambridge, in 1999. He is currently an Associate Pro-fessor in the Department of Electrical and Computer Engineering at the Uni-versity of Toronto, Toronto, ON, Canada. His research interests include analogand digital VLSI circuits, systems and algorithms for energy-efficient signalprocessing with applications to electrical, chemical and photonic sensory in-formation acquisition, biosensor arrays, brain-silicon interfaces, parallel signalprocessing, adaptive computing for pattern recognition, and implantable andwearable biomedical electronics.Dr. Genov received the Canadian Institutes of Health Research (CIHR) Next

GenerationAward in 2005, the Brian L. BargeAward for excellence in microsys-tems integration in 2008, the DALSA Corporation Award for excellence in mi-crosystems innovation in 2006 and 2009, and the Best Paper Award on sensorsand the Best Student PaperAward, both at the IEEE International Symposium onCircuits and Systems in 2009. He is an Associate Editor of IEEE TRANSACTIONSON BIOMEDICAL CIRCUITS AND SYSTEMS, IEEE TRANSACTIONS ON CIRCUITSAND SYSTEMS II: EXPRESS BRIEFS, and IEEE SIGNAL PROCESSING LETTERS.

Documents

468 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND …roman/professional/... · Index Terms—Biosensor, DNA, electrochemical sensor, fre-quency response analysis, impedance spectroscopy,